Application Performance

Individual Benchmark Scores

Super Pi

SuperPi is one of the most popular benchmarks with overclockers and tweakers. It has been used in world-record competitions since forever. It is a purely single-threaded CPU test that calculates Pi to a large number of digits—32 million for our testing. Released in 1995, it only supports x86 floating-point instructions and thus makes for a good test for single-threaded legacy application performance.

wPrime

While SuperPi focuses on calculating Pi, wPrime tackles another mathematical problem: finding prime numbers. It uses Newton's Method for that. One of the design goals for wPrime was to engineer it so that it can make the best use of all cores and threads available on a processor.

Rendering — Cinebench

Cinebench is one of the most popular modern CPU benchmarks because it is built around the renderer of Maxon's Cinema 4D software. Both AMD and Intel have been showing this performance test at various public events, making it almost an industry standard. Using Cinebench R23, we test both single-threaded and multi-threaded performance.

Rendering — Blender

Blender is one of the few professional-grade rendering programs out there that is both free and open source. That fact alone helped build a strong community around the software, making it a highly popular benchmark program due to its ease of use as well. For our testing, we're using the Blender "BMW 27" benchmark scene with Blender v2.92.

Rendering — Corona

Corona Renderer is a modern photorealistic renderer that's available for Autodesk 3ds Max and Cinema 4D. It delivers physically plausible and predictable output due to its realistic lighting algorithm, global illumination, and beautiful materials. Corona does not support GPU rendering, so CPU performance is very important for all its users. We do not use the horribly outdated Corona Benchmark tool, which doesn't support new architectures, but use the latest version 6.1.

Rendering — KeyShot

The standalone KeyShot rendering software features fast and efficient workflows that help you get realistic high-quality product shots in the shortest possible time frame. Real-time ray tracing, multi-core photon mapping, adaptive material sampling, and a dynamic lighting core provide high-quality images that update instantly even when interactively working on the scene. KeyShot 10 is optimized for usage on both CPUs and GPUs. We use the CPU renderer because the GPU renderer still has some limitations in terms of rendering capability.

Rendering — V-Ray

V-Ray is a world-leading 3D rendering software that uses global illumination, path tracing, photon mapping, and irradiance maps to achieve super-realistic render output. It has been used for CG in countless motion pictures and television series. V-Ray supports all major 3D applications, which makes it a great fit for any rendering pipeline. In this test, we're using the V-Ray 5 benchmark tool in CPU-only mode to get the number of "vsamples" that can be processed on a given hardware. Higher is better.

Game Development — Unreal Engine 4

Unreal Engine 4 is one of the leading multi-platform game engines in the industry. Not only advanced, it also has lots of features to help you get results quicker than with competing products—time is money. Before a game is shipped, the lengthy process of "light baking" has to be executed. It takes all static geometry and fixed light sources in the scene and precalculates lightmap textures for them, which results in a tremendous performance speed boost of the final game because these calculations no longer have to be performed in real time on the user's system. For our benchmarks, we generate baked light maps for a relatively simple scene—usually baking light takes several hours.

Software Development — Visual Studio C++

Microsoft Visual C++ is probably the most popular programming language for creating professional Windows applications. It's part of Microsoft's Visual Studio development suite, which has a long history and is widely accepted as the gold standard when it comes to IDEs. Compiling software is a relatively lengthy process that turns program code into the final executable, and programmers hate waiting for it to complete. We run a medium-sized application through the Visual Studio 2019 16.9 C++ compiler and linker, and use the resource compiler, too. The build is executed in "release" mode with all optimizations turned on and multi-processor compilation enabled.

Web Browsing — Google Octane

Google Octane tests web browser performance by running a suite of Javascript-based tests that represent typical use cases in today's dynamic, interactive web applications.

Web Browsing — Mozilla Kraken

Mozilla Kraken is similar to Octane in that it measures the execution time of Javascript, but uses a different set of tests based on the SunSpider benchmark. Test cases include audio processing, search algorithms, image filtering, JSON parsing, and cryptography.

Web Browsing — WebXPRT

WebXPRT 3 is a browser benchmark that measures the performance of typical web applications, like photo enhancement, media management using AI, stock option pricing, encryption, OCR, charting, and productivity. This is in contrast to our other two browser benchmarks which focus more on microbenchmarks, testing specific algorithms.

Machine Learning — Image Upscaling

Artificial Intelligence and Machine Learning have enabled us to create applications that are almost magical in their abilities. In our first AI benchmark we use Topaz Gigapixel AI to upscale a series of low resolution photos to higher resolution, while improving fine details at the same time.

Machine Learning — Image Classification

Teaching an AI to accurately recognize the contents of an image has always been one of the holy grails of Machine Learning research. Nero AI Photo Tagger is a consumer-oriented application that implements a simple way to use these algorithms. We let the software classify 2500 photos with tags like "car," "dog," or "flower."

Machine Learning — Tensorflow

Building an AI to solve complex problems first requires it to be trained through a large training data set that is evaluated repeatedly to generate a neural network that can later be put to work (also called inference). Google's Python-based Tensorflow is one of the most popular machine-learning software packages that supports both CPUs and GPUs. Setting up Tensorflow for the GPU is a bit complicated, so lots of algorithm development and training on small data sets still happens on the CPU. Training performance on the CPU can also beat the GPU when problem sizes exceed typical GPU memory capacities.

Physics Simulation

COMSOL is the gold standard for the accurate simulation of physics effects and one of the strongest competitors to Ansys. Where the software shines is in solving multiphysics problems that depend on several phenomena—you can simulate everything, including black holes. The modular aspect of COMSOL helps professionals customize the software to their desires, which can be run with a single module or a whole multitude of collaborating modules for more demanding tasks, as well as with two-way communication with external programs, such as CAD software.

Chemistry Simulation

NAMD is developed by the University of Illinois and helps researchers simulate the dynamic interactions between molecules and atoms. It is an established means of molecular dynamics simulation that scales from individual clusters of molecules to full-on chemical interaction in different phases. Parallel simulation modes help differentiate NAMD from its peers, allowing efficient problem solving of larger systems using the Charm++ parallel programming model as the base.

Microsoft Office

Microsoft's Office suite needs no introduction as it's probably the most widely used PC software on the planet, installed on every office computer no matter the industry. Our tests cover a wide range of editing and creation tasks in Word, PowerPoint, and Excel.

Image Editing — Adobe Photoshop

Adobe Photoshop has become the industry standard for photo and image processing. We run the latest Photoshop CC through a battery of typical editing tasks, like image resize, various blurs, sharpening, color and light adjustments, and image export. We also apply more complex actions like content-aware fill, lens correction, smart sharpen, select subject, and tilt shift.

Video Editing — Adobe Premiere Pro

Adobe Premiere Pro CC is the workhorse of the video production industry for creating high-quality content for film, TV, and the web. It can handle pretty much every recorded file format and supports workflows for editing Full HD, 4K, 8K, and virtual reality content. Unfortunately, most of Premiere Pro is single-threaded, and media encoding is highly GPU accelerated, so benchmarking "export" on the CPU makes little sense. For our testing, we're using the software's "object tracking" functionality, which automatically scans through a video to follow a specific person or object—this task does indeed use more than a core, but doesn't fully scale. A lot of memory is consumed and accessed in the process, over 10 GB for our test scene.

Create 3D Model from Photos

Creating 3D models is a tedious and complex task that takes time and requires experienced artists. It's thus the holy grail of 3D modeling to reconstruct a 3D model from just a series of photos. That's exactly what Photogrammetry does. This method is also used to reconstruct terrain geometry from photos taken by aerial drones.

Text Recognition OCR — Google Tesseract

Optical Character Recognition, or OCR, is the task of turning text in scanned images or photos into actual characters for archival, further processing, or editing. While most OCR software is single-threaded, Google's Tesseract engine can operate on multiple pages of a scanned document at once, spreading the load over several processor cores. The software, which is considered one of the most accurate open-source OCR packages available, automatically runs a spellcheck on the initial recognition results, which adds to the complexity of the workload.

Database — MySQL

More data is stored and processed today than ever before in human history. Database systems that manage storage and retrieval throughout large data sets are the backbone for this revolution. Whenever you interact with a website or other digital service, one database is almost always guaranteed to be involved in returning the results you are looking for. We benchmark the most popular database system, MySQL, in the TPC-C test, which simulates a large number of warehouses and their constantly changing inventory. The reported number is "transactions per second," so higher is better.

Java

The Java programming language is designed to be platform independent, highly scalable, and fault tolerant, which is why it's very popular for enterprise services that work with large amounts of data and many concurrent users. Our test suite consists of a large mix of individual Java benchmarks, some of them single-threaded, some that scale somewhat, and some that fully scale to as many cores as are available.

Compression — WinRAR

Data is compressed almost all the time when it moves over the wire to reduce download time and transfer sizes. WinRAR uses a more advanced compression algorithm than classic ZIP, which is why we chose it for this test. It's also able to scale across multiple processor cores.

Compression — 7-Zip

Another popular compression software is 7-Zip, which includes a benchmark that measures the integer instruction rate (MIPS) using the ZIP algorithm. It makes good use of multiple threads, when available.

Encryption — AES

Encryption is the foundation for secure communication on the internet. The AES encryption standard is one of the most popular encryption algorithm these days because of its simplicity and resistance to attacks. Unlike other encryption methods, AES is symmetric, which means the same key is used for encrypting and decrypting data. Being able to quickly encrypt and decrypt information is important, which is the reason why modern processors have an instruction set extension called "AES-NI" that accelerates these operations.

Encryption — SHA3

"SHA" is technically not encryption, but hashing. A cryptographic hash function is a mathematical algorithm that creates a fingerprint value for existing data. This is a one-way function, which is practically impossible to invert. Another requirement for a good hashing algorithm is that you can't create a collision: craft a message that yields a specific hash value. The SHA3 algorithm achieves all that and is also often used to authenticate transmitted data which possibly was encrypted with AES first.

Media Encoding — AV1

Nowadays, all video that is consumed, whether on TV, physical media, or streamed over the internet, is compressed using various codecs. AV1 is a next-generation video codec that is open and royalty-free, unlike other recent new video compression algorithms. Major industry players like YouTube, Netflix, and Facebook stand behind AV1 and are gradually rolling out support for the new format. Compared to older algorithms, like H.264, the compression rate is much better, or you can achieve better image quality for a given bitrate. We use the SVT-AV1 encoder with a 4K video source to compress to AV1.

Media Encoding — H.265 / HEVC

Our second video-encoding test uses the H.265 codec, which is also known as HEVC. We compress a 4K video using the latest version of the X265 encoder with 8-bit color depth, preset "slow," and a quality setting of crf 20.

Media Encoding — H.264 / AVC

H.264, also called AVC, is a slightly older compression format, though probably the most widely used these days because it is well supported in even older hardware. We compress the same video as in the H.265 test using the X264 encoding software, with the preset "slower" and crf 20.

Media Encoding — MP3

MP3 revolutionized the music industry like no other technology. Introduced in the 90s, it enabled massive reduction in audio-file sizes without noticeable impact on sound quality. This made music downloads, and ultimately streaming, a feasible method of content delivery over the Internet. For our benchmark, we convert a 2.5 hour-long 44.1 kHz Stereo recording to a variable bitrate MP3 file. MP3 encoding is a single-threaded process.